I think this limitation could be overcome with the right hardware.
For example imagine a spinning display like those of the article but somehow tuned so that they are only visible when exactly head on. In that case, you know where the observer is: right in line with the screen. So you can have backface culling; as the display spins you render all 360 (or however many) viewpoints.
Now granted, this doesn't deal with how high or low the observer is. We'd need to find another solution for that.
This would be a poor man's "lightfield" display: as you move left or right you see a different perspEctive, just like you would if it were a physical object on the table instead of the spinning screen contraption.
For example imagine a spinning display like those of the article but somehow tuned so that they are only visible when exactly head on. In that case, you know where the observer is: right in line with the screen. So you can have backface culling; as the display spins you render all 360 (or however many) viewpoints.
Now granted, this doesn't deal with how high or low the observer is. We'd need to find another solution for that.